What investors are missing about AI

The rise of generative artificial intelligence has stoked excitement and optimism among business leaders and investors alike. But sentiment toward AI has recently turned much more skeptical, and many are now wondering whether AI’s potential has been overstated.  

Eric Sheridan, the senior research analyst covering the US Internet sector within Goldman Sachs Research, believes that AI has the potential to be a significant driver of tech companies’ financial results. But he acknowledges that “much more is unknown than known” right now. We interviewed him on the sidelines of Goldman Sachs’ Communacopia + Technology conference.

Is the promise of generative AI overblown?

I don’t think the promise is overblown, but AI is still squarely in the build mode. And if you go back and look at almost every computing cycle we’ve been through – and I have been a combination of investor and analyst through a number of them – there’s a product that captures imagination, there’s a build cycle, and then there’s always a bit of disillusionment on the time between the build cycle and applications that have utility. So if you listen to the companies broadly at the conference the last couple days, we’re moving deeper into the build cycle, which is putting an upward pressure on capital expenditures. But we haven’t yet had the unlock moment on either the consumer side or the enterprise side. 99% of your computing habits are the same today as they were two years ago, even though we’ve had AI in the public domain for the last 19 months. But over time, typically those computing cycles play out and those habits change.  

From a capital markets perspective, would you characterize this as a moment of investor impatience?

Investors generally want visibility and linearity. And with any computing cycle, there are points in time that we’ve called “troughs of disillusionment” where it’s not quite as linear, and it’s not quite as visible. We know there was a big tick up in capital expenditures around AI in 2024 compared to 2023, and we’re on an upward trajectory. We could be entering a period where there’s extended and outsized volatility, because of a lack of linearity or visibility around where technology is going over the medium to long term. 

As a research group, what we tell people is: Betting on the long term in technology seems to have worked out over very long periods of time, but you do need some patience.  

What happens if the application just don’t emerge, and AI proves incapable of profoundly changing businesses and the economy?

Well, the spending to date has been concentrated among a handful of very large companies. There are, in my view, components of both offense and defense in what those companies are doing. Eventually investors will call for a return on that spend. At least through 2025, these companies have all said that it’s better to overinvest than underinvest, so that they don’t miss out on the opportunity. But this idea that they will spend irrespective of the return profile is not what I hear from the companies I cover. 

There’s also a debate about whether AI will benefit the mega-cap incumbents, or if it will displace them.

One of the key things to keep in mind with respect to AI is that you need a lot of capital, you need a lot of engineering talent, and you need a lot of data. That creates enormous barriers to entry. So even some of the most interesting private companies in the landscape have found their way to partnerships with some of the incumbents just because of the sheer scale of capital, the sheer need for engineering talent, and then the data required to scale these models. Keep in mind that a lot of the large incumbents have pretty big first-party walled-garden data pools, which can be very important in training and improving these models. 

But ultimately, if AI does prove to be a massive driver of the economy, how much of that value will be captured by tech companies?

One area where it will be captured is compute needs. You arguably would take some element of what you pay for work today, redeploy that money into compute, and then the compute would handle the productivity dynamic. We’ve heard people on stage talk about “the virtual worker.” Well, you don’t have to pay the virtual worker, but there are compute costs underneath it.  

Do we generally understand the types of impacts that AI will have at this point? Or are investors and business leaders still making educated guesses?

I think much more is unknown than known. Across all the enterprises we talk to, we generally hear a lot of “I’m testing,” “I’m learning,” “I’m seeing what can be built.” So I think if someone says, “I know exactly how AI is going to play out over the next ten years,” I wouldn’t put a lot of weight on that.  As a research department, we’ve tried to talk about probabilistic outcomes, and how the landscape can evolve. 

Is there anything that investors consistently miss about the AI story?

Within the consumer technology space, I’ve found that the advantages of distribution are always underestimated by investors. If you have an existing base of billions of users, that is an incredible advantage. And if you then iterate by putting a little bit more consumer-facing AI into those products, that might not be where the bleeding-edge tech is, but it’s how the average consumer will experience it.

 

This article is being provided for educational purposes only. The information contained in this article does not constitute a recommendation from any Goldman Sachs entity to the recipient, and Goldman Sachs is not providing any financial, economic, legal, investment, accounting, or tax advice through this article or to its recipient. Neither Goldman Sachs nor any of its affiliates makes any representation or warranty, express or implied, as to the accuracy or completeness of the statements or any information contained in this article and any liability therefore (including in respect of direct, indirect, or consequential loss or damage) is expressly disclaimed.